Artificial intelligence and animals: machine learning used to decode dogs’ brains

A team of neuroscientists fromPrinceton University used machine learning to decode the visual processing that takes place in the ghost of two dogs. Using magnetic resonance imaging techniques, scientists have discovered a fascinating difference between the perception of dogs and humans: Dogs are much more attentive to actions than to who or what they are doing.

The study was conducted on two dogs, Daisy and Bhubo. Thanks to the use of some training techniques, it was possible to keep the two dogs in an MRI machine without restrictions. During the sessions, Daisy and Bhubo were shown videos with “dog-specific” content. The Princeton research team shot three 30-minute videos using a stabilizer and a selfie stick. Among the images recorded are, for example, dogs running and people handling dogs, vehicles passing by on the street, a deer crossing a path, a cat in a house and dogs being carried on a leash.

How did the experiment go?

Daisy and Bhubo, seated comfortably like in the cinema, got the filmed in three 30 minute sessions each, for a total of 90 minutes. The researchers segmented some of the information in the data encoding videos: They categorized objects (for example, dogs, people, vehicles or other animals) or actions (such as sniffing, eating or playing). This information, as well as the brain activity of the two dogs, was recorded by an in vivo imaging system. IVISwho manages to combine 2D and 3D optical tomography to map brain activity.

The researchers were then able to scan Daisy and Bhubo’s brains while they were awake, alert and comfortable in the MRI machine, looking at images created especially for them. The research team also decided to involve two people in the study, to analyze the differences in a dog’s and a person’s mental processes. They showed them the same videos as IVIS recorded the data. Artificial intelligence was able to map human brain data with 99% accuracy, for both object and action classification.

In dogs, there was less precision: Ivis didn’t get good results in object classification, but for actions, artificial intelligence mapped visual and brain activity with an accuracy of between 75 and 88%.

The results of the study

The results of this preliminary study they therefore tell us that humans are much more object-oriented, while dogs seem to be less interested in who or what they see and more interested in the action taking place. This is important information for cognitive studies in animals, as it shows what their brain’s priority is when it comes to vision. The elements this study provides us confirm something that is partly known: dogs show significant differences in the way they perceive the world compared to humans.

It is known that dogs do not have the same visual skills as we do, but for example only distinguish shades of the blue and yellow color spectrum. This is in contrast to a greater density of visual receptors that are sensitive to movement. This may be due to the fact that ancestral-level dogs need to be more aware of the threats in their environment than humans.

This study, although conducted on a very small sample (just two dogs), provides important insights for future research. For example, it may be interesting to do further, more detailed research into the visual perception of dogs and possibly other animals. Certainly noteworthy was the fact that it was possible to map the brain activity of the dogs, especially without limitations.

The research is published in the Journal of Visualized Experiments.